From: Mel Gorman <mgorman@techsingularity.net>
To: Ingo Molnar <mingo@kernel.org>
Cc: Aubrey Li <aubrey.intel@gmail.com>,
Julien Desfossez <jdesfossez@digitalocean.com>,
Vineeth Remanan Pillai <vpillai@digitalocean.com>,
Nishanth Aravamudan <naravamudan@digitalocean.com>,
Peter Zijlstra <peterz@infradead.org>,
Tim Chen <tim.c.chen@linux.intel.com>,
Thomas Gleixner <tglx@linutronix.de>,
Paul Turner <pjt@google.com>,
Linus Torvalds <torvalds@linux-foundation.org>,
Linux List Kernel Mailing <linux-kernel@vger.kernel.org>,
Subhra Mazumdar <subhra.mazumdar@oracle.com>,
Fr?d?ric Weisbecker <fweisbec@gmail.com>,
Kees Cook <keescook@chromium.org>, Greg Kerr <kerrnel@google.com>,
Phil Auld <pauld@redhat.com>, Aaron Lu <aaron.lwe@gmail.com>,
Valentin Schneider <valentin.schneider@arm.com>,
Pawan Gupta <pawan.kumar.gupta@linux.intel.com>,
Paolo Bonzini <pbonzini@redhat.com>,
Jiri Kosina <jkosina@suse.cz>
Subject: Re: [RFC PATCH v2 00/17] Core scheduling v2
Date: Fri, 26 Apr 2019 11:43:28 +0100 [thread overview]
Message-ID: <20190426104328.GA18914@techsingularity.net> (raw)
In-Reply-To: <20190426084222.GC126896@gmail.com>
On Fri, Apr 26, 2019 at 10:42:22AM +0200, Ingo Molnar wrote:
> > It should, but it's not perfect. For example, wake_affine_idle does not
> > take sibling activity into account even though select_idle_sibling *may*
> > take it into account. Even select_idle_sibling in its fast path may use
> > an SMT sibling instead of searching.
> >
> > There are also potential side-effects with cpuidle. Some workloads
> > migration around the socket as they are communicating because of how the
> > search for an idle CPU works. With SMT on, there is potentially a longer
> > opportunity for a core to reach a deep c-state and incur a bigger wakeup
> > latency. This is a very weak theory but I've seen cases where latency
> > sensitive workloads with only two communicating tasks are affected by
> > CPUs reaching low c-states due to migrations.
> >
> > > Clearly it doesn't.
> > >
> >
> > It's more that it's best effort to wakeup quickly instead of being perfect
> > by using an expensive search every time.
>
> Yeah, but your numbers suggest that for *most* not heavily interacting
> under-utilized CPU bound workloads we hurt in the 5-10% range compared to
> no-SMT - more in some cases.
>
Indeed, it was higher than expected and we can't even use the excuse that
more resources are available to a single logical CPU as the scheduler is
meant to keep them apart.
> So we avoid a maybe 0.1% scheduler placement overhead but inflict 5-10%
> harm on the workload, and also blow up stddev by randomly co-scheduling
> two tasks on the same physical core? Not a good trade-off.
>
> I really think we should implement a relatively strict physical core
> placement policy in the under-utilized case, and resist any attempts to
> weaken this for special workloads that ping-pong quickly and benefit from
> sharing the same physical core.
>
It's worth a shot at least. Changes should mostly be in the wake_affine
path for most loads of interest.
> I.e. as long as load is kept below ~50% the SMT and !SMT benchmark
> results and stddev numbers should match up. (With a bit of a leewy if the
> workload gets near to 50% or occasionally goes above it.)
>
> There's absolutely no excluse for these numbers at 30-40% load levels I
> think.
>
Agreed. I'll put it on the todo list but there is no way I'll get to it
in the short term due to LSF/MM. Minimally I'll put some thought into
tooling to track how often siblings are used with some reporting on when a
sibling was used when there was an idle core available. That'll at least
quantify the problem and verify the hypothesis.
--
Mel Gorman
SUSE Labs
next prev parent reply other threads:[~2019-04-26 10:43 UTC|newest]
Thread overview: 109+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-04-23 16:18 [RFC PATCH v2 00/17] Core scheduling v2 Vineeth Remanan Pillai
2019-04-23 16:18 ` [RFC PATCH v2 01/17] stop_machine: Fix stop_cpus_in_progress ordering Vineeth Remanan Pillai
2019-04-23 16:18 ` [RFC PATCH v2 02/17] sched: Fix kerneldoc comment for ia64_set_curr_task Vineeth Remanan Pillai
2019-04-23 16:18 ` [RFC PATCH v2 03/17] sched: Wrap rq::lock access Vineeth Remanan Pillai
2019-04-23 16:18 ` [RFC PATCH v2 04/17] sched/{rt,deadline}: Fix set_next_task vs pick_next_task Vineeth Remanan Pillai
2019-04-23 16:18 ` [RFC PATCH v2 05/17] sched: Add task_struct pointer to sched_class::set_curr_task Vineeth Remanan Pillai
2019-04-23 16:18 ` [RFC PATCH v2 06/17] sched/fair: Export newidle_balance() Vineeth Remanan Pillai
2019-04-23 16:18 ` [RFC PATCH v2 07/17] sched: Allow put_prev_task() to drop rq->lock Vineeth Remanan Pillai
2019-04-23 16:18 ` [RFC PATCH v2 08/17] sched: Rework pick_next_task() slow-path Vineeth Remanan Pillai
2019-04-23 16:18 ` [RFC PATCH v2 09/17] sched: Introduce sched_class::pick_task() Vineeth Remanan Pillai
2019-04-26 14:02 ` Peter Zijlstra
2019-04-26 16:10 ` Vineeth Remanan Pillai
2019-04-29 5:38 ` Aaron Lu
2019-04-23 16:18 ` [RFC PATCH v2 10/17] sched: Core-wide rq->lock Vineeth Remanan Pillai
2019-04-23 16:18 ` [RFC PATCH v2 11/17] sched: Basic tracking of matching tasks Vineeth Remanan Pillai
2019-04-24 0:08 ` Tim Chen
2019-04-24 20:43 ` Vineeth Remanan Pillai
2019-04-24 22:12 ` Tim Chen
2019-04-25 14:35 ` Phil Auld
2019-05-22 19:52 ` Vineeth Remanan Pillai
2019-04-24 0:17 ` Tim Chen
2019-04-24 20:43 ` Vineeth Remanan Pillai
2019-04-29 3:36 ` Aaron Lu
2019-05-10 13:06 ` Peter Zijlstra
2019-04-29 6:15 ` Aaron Lu
2019-05-01 23:27 ` Tim Chen
2019-05-03 0:06 ` Tim Chen
2019-05-08 15:49 ` Aubrey Li
2019-05-08 18:19 ` Subhra Mazumdar
2019-05-08 18:37 ` Subhra Mazumdar
2019-05-09 0:01 ` Aubrey Li
2019-05-09 0:25 ` Subhra Mazumdar
2019-05-09 1:38 ` Aubrey Li
2019-05-09 2:14 ` Subhra Mazumdar
2019-05-09 15:10 ` Aubrey Li
2019-05-09 17:50 ` Subhra Mazumdar
2019-05-10 0:09 ` Tim Chen
2019-04-23 16:18 ` [RFC PATCH v2 12/17] sched: A quick and dirty cgroup tagging interface Vineeth Remanan Pillai
2019-04-25 14:26 ` Phil Auld
2019-04-26 14:13 ` Peter Zijlstra
2019-04-26 14:19 ` Phil Auld
2019-05-10 15:12 ` Julien Desfossez
2019-04-23 16:18 ` [RFC PATCH v2 13/17] sched: Add core wide task selection and scheduling Vineeth Remanan Pillai
2019-04-29 7:13 ` Aaron Lu
2019-05-18 15:37 ` Aubrey Li
2019-05-20 13:04 ` Phil Auld
2019-05-20 14:04 ` Vineeth Pillai
2019-05-21 8:19 ` Aubrey Li
2019-05-21 13:24 ` Vineeth Pillai
2019-04-23 16:18 ` [RFC PATCH v2 14/17] sched/fair: Add a few assertions Vineeth Remanan Pillai
2019-04-23 16:18 ` [RFC PATCH v2 15/17] sched: Trivial forced-newidle balancer Vineeth Remanan Pillai
2019-04-23 23:46 ` Aubrey Li
2019-04-24 14:03 ` Vineeth Remanan Pillai
2019-04-24 14:05 ` Vineeth Remanan Pillai
2019-04-23 16:18 ` [RFC PATCH v2 16/17] sched: Wake up sibling if it has something to run Vineeth Remanan Pillai
2019-04-26 15:03 ` Peter Zijlstra
2019-04-29 12:36 ` Julien Desfossez
2019-04-23 16:18 ` [RFC PATCH v2 17/17] sched: Debug bits Vineeth Remanan Pillai
2019-05-17 17:18 ` Aubrey Li
2019-04-23 18:02 ` [RFC PATCH v2 00/17] Core scheduling v2 Phil Auld
2019-04-23 18:45 ` Vineeth Remanan Pillai
2019-04-29 3:53 ` Aaron Lu
2019-05-06 19:39 ` Julien Desfossez
2019-05-08 2:30 ` Aaron Lu
2019-05-08 17:49 ` Julien Desfossez
2019-05-09 2:11 ` Aaron Lu
2019-05-15 21:36 ` Vineeth Remanan Pillai
2019-04-23 23:25 ` Aubrey Li
2019-04-24 11:19 ` Vineeth Remanan Pillai
2019-05-15 21:39 ` Vineeth Remanan Pillai
2019-04-24 13:13 ` Aubrey Li
2019-04-24 14:00 ` Julien Desfossez
2019-04-25 3:15 ` Aubrey Li
2019-04-25 9:55 ` Ingo Molnar
2019-04-25 14:46 ` Mel Gorman
2019-04-25 18:53 ` Ingo Molnar
2019-04-25 18:59 ` Thomas Gleixner
2019-04-25 19:34 ` Ingo Molnar
2019-04-25 21:31 ` Mel Gorman
2019-04-26 8:42 ` Ingo Molnar
2019-04-26 10:43 ` Mel Gorman [this message]
2019-04-26 18:37 ` Subhra Mazumdar
2019-04-26 19:49 ` Mel Gorman
2019-04-26 9:45 ` Ingo Molnar
2019-04-26 10:19 ` Mel Gorman
2019-04-27 9:06 ` Ingo Molnar
2019-04-26 9:51 ` Ingo Molnar
2019-04-26 14:15 ` Phil Auld
2019-04-26 2:18 ` Aubrey Li
2019-04-26 9:51 ` Ingo Molnar
2019-04-27 3:51 ` Aubrey Li
2019-04-27 9:17 ` Ingo Molnar
2019-04-27 14:04 ` Aubrey Li
2019-04-27 14:21 ` Ingo Molnar
2019-04-27 15:54 ` Aubrey Li
2019-04-28 9:33 ` Ingo Molnar
2019-04-28 10:29 ` Aubrey Li
2019-04-28 12:17 ` Ingo Molnar
2019-04-29 2:17 ` Li, Aubrey
2019-04-29 6:14 ` Ingo Molnar
2019-04-29 13:25 ` Li, Aubrey
2019-04-29 15:39 ` Phil Auld
2019-04-30 1:24 ` Aubrey Li
2019-04-29 16:00 ` Ingo Molnar
2019-04-30 1:34 ` Aubrey Li
2019-04-30 4:42 ` Ingo Molnar
2019-05-18 0:58 ` Li, Aubrey
2019-05-18 1:08 ` Li, Aubrey
2019-04-25 14:36 ` Julien Desfossez
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20190426104328.GA18914@techsingularity.net \
--to=mgorman@techsingularity.net \
--cc=aaron.lwe@gmail.com \
--cc=aubrey.intel@gmail.com \
--cc=fweisbec@gmail.com \
--cc=jdesfossez@digitalocean.com \
--cc=jkosina@suse.cz \
--cc=keescook@chromium.org \
--cc=kerrnel@google.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@kernel.org \
--cc=naravamudan@digitalocean.com \
--cc=pauld@redhat.com \
--cc=pawan.kumar.gupta@linux.intel.com \
--cc=pbonzini@redhat.com \
--cc=peterz@infradead.org \
--cc=pjt@google.com \
--cc=subhra.mazumdar@oracle.com \
--cc=tglx@linutronix.de \
--cc=tim.c.chen@linux.intel.com \
--cc=torvalds@linux-foundation.org \
--cc=valentin.schneider@arm.com \
--cc=vpillai@digitalocean.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).